At present, voice communication using a network such as the Internet has been actively conducted by the use of the VoIP technology.
In communication over a network such as the Internet in which the communication quality is not assured, a packet loss that a packet is lost during transmission frequently causes a phenomenon that a part of voice data, which is supposed to be received in a time series under normal circumstances, is lost. When a voice loss occurs, if the voice data is decoded as it is, voice is frequently interrupted to degrade the voice quality. A technology disclosed in non-patent document 1 to be described below has been already known as a method for compensating this degradation.
In this method, a sampling frequency of 8 kHz is described for voice and the occurrence of a voice loss is monitored for each voice frame (packet) of a decoding processing unit and every time a voice loss occurs, compensation processing is performed. Because voice data after decoding a series of coding data is stored in an internal memory or the like, when the voice loss occurs, a period near a portion where the voice loss occurs is obtained on the basis of the voice data read from the internal memory. Voice data is taken out of the internal memory and interpolation is performed to a frame in which voice data need to be interpolated because of the voice loss so that the starting phase of the frame matches with the ending phase of a frame immediately preceding the frame to thereby secure continuity of a waveform period.
Meanwhile, technologies described in non-patent documents 2 and 3 to be described below are known as a method of voice communication over a network.
In the technology described in the non-patent document 2, the amplitude values of sampling voice are quantized just as they are. In the technology described in the non-patent document 3 is used a difference quantization method for quantizing the amount of change in an amplitude value between sampling points.
In a difference quantization method, a difference in a voice signal between sampling points is obtained by an encoder for encoding voice and is quantized to make a difference signal and the difference signal is transmitted. A decoder for receiving the transmitted difference signal decodes the received difference signal to an original voice signal. In this difference quantization method, the encoder and the decoder have common internal variables used for computing and converting the difference signal and the original signal. Hence, the internal variables in the encoder and the decoder are always updated for the time period during which the encoder and the decoder are operating.    Non-patent document 1: ITU-T Recommendation G.711 Appendix I    Non-patent document 2: ITU-T Recommendation G.711    Non-patent document 3: ITU-T Recommendation G.726